Launching today
YABAI
Real Japanese slang, one giant word at a time
8 followers
Real Japanese slang, one giant word at a time
8 followers
Textbooks teach Japanese that stops at the classroom door. YABAI is the other half: 308 real slang, internet and fandom words, each on a card built around giant Japanese typography. Every word has a meaning, a real example, a vibe check (safe to use? friends only? careful?) and a story — and when the origin is uncertain, it says so instead of inventing one. Built end to end with AI agents in Claude Code. 13 languages, works offline. No streaks, no tests, no XP.
Products used by YABAI
Explore the tech stack and tools that power YABAI. See what products YABAI uses for development, design, marketing, analytics, and more.
Design & Creative 2
Design & Creative 2
RemotionCreate videos programmatically
5.0 (56 reviews)
I needed a series, not one video — ten vertical shorts so far, one per Japanese word, eventually a hundred. In Remotion each video is a data file plus shared React components, so adding a word is a data change and a render, not a new edit session. Being able to re-render all of them after a single copy change is what made the series viable for one person. Narration is generated locally and the captions come from the same data, so audio and subtitles cannot drift apart.
FlutterBuild apps for iOS, Android, web, macOS, Windows, & Linux
5.0 (128 reviews)
Also considered:
React NativeThis app is essentially a typography engine: giant Japanese type, a paper-grain texture drawn at runtime from a seeded random, and motion that has to land identically everywhere. Flutter draws every pixel itself, so a card looks right on a cheap Android phone and on a tablet without chasing platform text-rendering differences. My layout test matrix (11 screen heights x 5 text scales, including Thai, which wraps without spaces) was only practical because the rendering is deterministic.
Engineering & Development 1
Engineering & Development 1
Claude CodeAnthropic’s deep-context AI coder
5.0 (701 reviews)
Also considered:
openai-codexI ran the entire project inside it: the Flutter app, a 308-word dictionary written in Japanese and translated into 13 languages, the store assets, and the promo videos. What mattered most was being able to fan work out to subagents with clearly separated file ownership and then review their output myself, instead of babysitting one long thread. I used Codex alongside it for read-only translation audits and the two were complementary — Codex was good at sweeping 13 languages for errors in parallel, Claude Code at holding the design intent steady across months of work.